What NVIDIA Actually Got Wrong with the DLSS 5 Reveal

Nobody, except for a very select few who had already checked out a demo, was expecting NVIDIA CEO Jensen Huang to unveil DLSS 5 during his GTC 2026 keynote last week. NVIDIA had already countered AMD's FSR Redstone at CES 2026, where it announced and released a second-generation transformer model for its Super Resolution upscaler, which adopted the DLSS 4.5 label and was quickly judged the best upscaler available. Two months later, they stunned the world - for better or worse - with a new version of DLSS that did not focus on improving frame rates or ray tracing performance, but instead sought to deliver photorealistic lighting and shading with the power of AI.

Huang, now one of the richest men in the world after NVIDIA's swift climb to the top of the world's most valuable publicly traded companies powered by AI demand, said that "just as GeForce brought AI to the world, AI is now going to go back and revolutionize how computer graphics is done altogether", calling DLSS 5 the future and next-generation of graphics technology. However, the gaming community's reaction was very mixed. While the first tech journalists who checked out the demo were largely impressed, the DLSS 5 reveal was criticized by many gamers, as well as some developers and modders, some of whom we've recently interviewed on Wccftech.

Personally, I disagree with most of the criticism levied against the technology. However, in hindsight, it seems clear to me that NVIDIA made a few critical mistakes when it comes to the reveal itself, ultimately influencing the public's first impression of this new version of DLSS.

That Grace Thumbnail

It's incredible how one screenshot has singlehandedly hijacked the whole DLSS 5 conversation. Yes, I'm referring to the Resident Evil Requiem comparison showing co-protagonist Grace Ashcroft from a scene at the very beginning of the game. NVIDIA chose that one as the key image to showcase the massive visual leap the new technology delivers. It was even picked as the featured image for the official announcement blog post.

And demonstrate it did, but not in a good way. While it may have impressed non-gamers at GTC, it was immediately singled out by gamers as proof that NVIDIA was just adding an AI-powered "beauty filter" to games that would, however, ruin the original game's artistic and narrative intent. They even went on to elaborate that Grace wouldn't really put any makeup on in this case, because she is on her way to investigate the circumstances of her mother's death, hardly an appropriate occasion to dress up.

NVIDIA failed to fully grasp the strong emotional attachment that millions of gamers who had just finished and loved Resident Evil Requiem had formed with the character's existing look. What's worse is that, if you actually browse the entire gallery of official DLSS 5 screenshots, there is another comparison dedicated to Grace that shows a much closer rendition of the character.

2 of 9

Whereas that infamous thumbnail was rightfully criticized by all, this other, far less discussed comparison demonstrates how the technology can largely preserve a game character's original facial structure while still greatly improving its lighting and shading. The image choice suggests that someone at NVIDIA wanted to maximize the wow factor with that specific thumbnail choice, and in doing so, essentially condemned DLSS 5 to a fierce backlash.

Showing Off the Tech Way Too Early

As I noted earlier, nobody was even thinking about a new DLSS, yet NVIDIA was all too eager to tell the whole world about it despite being far from ready. They showed it off over half a year before the planned Fall 2026 launch, at a time when it still required two GeForce RTX 5090 graphics cards to run.

That never happened before with any prior DLSS reveal, and it's a clear tell that the reveal came far too early in the tech's development cycle. This ties into the previous mistake: Grace's first comparison shows that NVIDIA still has tuning left to do to ensure the DLSS 5 model is far more conservative when handling faces. They could have either waited until a later date, or picked those screenshots - and there are many, including renderings of Liverpool FC captain Virgil van Dijk from the game EA Sports FC - where the model already keeps close to the original while delivering a series of substantial lighting improvements.

2 of 9

They did neither, and they suffered for it. Someone perhaps wanted a cool new technology to show off at GTC, but the price of this rushed reveal turned out to be high. It would have been wiser to wait until DLSS 5 was more stable, closer to release, and capable of running on a single GeForce RTX GPU.

Neural Rendering: Yes or No?

All that being said, I do believe much of the hate is unwarranted. First of all, neural rendering is not an option: it's the only remaining venue for improvement. Moore's Law is dead, and the remaining silicon-derived improvements won't be enough on their own to achieve true photorealism in real-time rendering. Many people say that games look nearly like CGI films nowadays, but that really isn't true. Real-time rendering still lags far behind in both lighting and shading, and that applies to path traced games as well.

Neural rendering techniques are the only way to bridge a gap that could otherwise persist for years, if not decades. And it's not like NVIDIA didn't tell us they were going in this direction; a couple of years ago, VP of Applied Deep Learning Research and DLSS father Bryan Catanzaro famously said:

I do think that let's say DLSS 10 in the far future is going to be a completely neural rendering system that interfaces with a game engine in different ways, and because of that, it's going to be more immersive and more beautiful.

DLSS 5 doesn't integrate with a game engine yet (which was one of the side criticisms), but that certainly sounds like the next step for the technology and would ensure better harmony with a game's original style. Indeed, Samson developer Liquid Swords told Wccftech that it's basically a requirement for them to have it fully integrated into the character production pipeline.

At CES 2026, Huang had once again teased that the future of graphics would be powered by neural rendering. Now, it's official, albeit earlier than expected, and perhaps understandably still a bit rough.

A Familiar Pattern: "Fake" Pixels, "Fake" Frames, "Fake" Faces?

This is far from the first NVIDIA DLSS controversy. Ever since NVIDIA introduced the first version of its Deep Learning Super Sampling suite, it has been the target of criticism. With DLSS 1.0, those not fond of NVIDIA cried about the "fake pixels" (DLSS renders the game at a lower internal resolution, depending on the user-chosen quality mode, and reconstructs the rest with AI). By the time DLSS 2.0 shipped with a significant improvement in image quality, it was clear that fake pixels were necessary, and AMD developed its own FidelityFX Super Resolution, while Intel worked on its Xe Super Sampling.

Then, with DLSS 3.0, NVIDIA went one step further and introduced Frame Generation, also known disparagingly as "fake frames". Again, there was strong outrage from purists, who could not bear the thought of having to deal with fully AI-generated frames between actually rendered ones. Guess what? AMD and Intel followed suit here, too, eventually introducing their own machine learning powered versions. Even Sony has recently admitted it will add an ML frame generation to its PlayStation consoles in the future.

As a pioneer, NVIDIA has shown time and again the way forward for rendering in the industry, even if it meant exposing the first iteration of a new technology to widespread criticism. DLSS 5 is no different; I have little doubt that AMD and Intel will take the same path in due time. Yes, NVIDIA made some glaring mistakes with this reveal, which I have outlined above. Yes, they need to improve the model before its full release, and possibly expand the number of tuning knobs available to developers and look into tighter integration with game engines.

I would even recommend delaying the actual release of the technology beyond Fall, if at all necessary. I doubt there's any risk of anyone getting a similar neural rendering tech out before NVIDIA, and it's important that the next time DLSS 5 is shown, most of the rational critiques (not the ideological, preconceived ones against any form of AI) are addressed properly.

As pointed out by tech journalist Ryan Shrout, one of the first to check out the demo in-person, forget about the faces for a minute: DLSS 5 lighting provides noticeable, next-generation-like improvements to 3D scenes, making objects, environments, and even things like water and foliage look a lot more life-like.

That's why throwing the entire thing into the bin, as the most extreme knee-jerk social media reactions would have NVIDIA do, just because the model still needs to be tuned to ensure faces always retain their original look, would be incredibly naive and short-sighted.

Follow Wccftech on Google to get more of our news coverage in your feeds.